Neurocomputing ] (]]]]) ]]]–]]] Neuromuscular control of reactive behaviors for undulatory robots M. Sfakiotakis, D.P. Tsakiris à Institute of Computer Science-FORTH, Vassilika Vouton, P.O. Box 1385, GR-71110 Heraklion, Greece Abstract Undulatory locomotion is studied as a biological paradigm of versatile body morphology and effective motion control, adaptable to a large variety of unstructured and tortuous environmental conditions. Computational models of undulatory locomotion have been developed, and validated on a series of robotic prototypes propelling themselves on sand. The present paper explores in simulation neuromuscular motion control for these undulatory robot models, based on biomimetic central pattern generators and on information from distributed distance sensors. This leads to reactive control schemes, which achieve (i) traversal of corridor-like environments, and (ii) formation control for swarms of undulatory robots. r 2006 Elsevier B.V. All rights reserved. Keywords: Neural control; Central pattern generators; Biomimetic robotics; Undulatory locomotion; Sensors; Closed-loop control; Polychaete annelids 1. Introduction Motion control is one of the most significant problems for emerging robotic applications dealing with locomotion in unstructured environments, which range from endo- scopy to planetary exploration [4,13,31,34]. Drawing inspiration from biology, where this problem has been effectively addressed by the evolutionary process, can help the design of agile robots able to adapt robustly to a variety of environmental conditions. Such an intriguing biological paradigm is offered by the polychaete annelid worms, whose locomotion is characterized by the combination of a unique form of tail-to-head body undulations, with the rowing-like action of numerous lateral appendages, called parapodia, distributed along their segmented body [6,11]. This provides the worms with distinctive locomotory modes, increasing their swimming, terrain traversing and burrowing capabilities over water, sand, mud and sedi- ment. These locomotory modes have been modeled computationally and validated via robotic prototypes propelling themselves on sand [30,31,34]. Undulatory locomotion in annelids and other organisms, as well as in robots, is achieved through appropriate coupling of internal shape changes (typically a traveling body wave) to external motion constrains (typically frictional forces from the interaction with the locomotion environment). Evidence exists [21,9,27] that motion control of the annelid undulatory locomotion is based on central pattern generators (CPGs), which are neuronal circuits able to produce rhythmic motor patterns in an organism, even in the absence of sensory input or of input from higher cognitive elements; such inputs may modulate the rhythmic activity of the CPG [23,2]. The typical morphology of annelids [11,6,27] hints at a sequential, modular and distributed sensing and control architecture, not unlike that of the CPG controlling the undulatory swimming of lamprey eels, which has been extensively studied in neurobiology [27,12] and modeled at various levels of detail [10,12,14,24,18,19,28]. From an engineering view- point, interest in CPG-based locomotion controllers, especially for undulatory mechanisms, stems not only from their elegance, but also from their potential to lead to distributed, fault-tolerant and robust motion control architectures [4,30,26,33,7,22,15,16]. The literature on undulatory robotic systems has mainly focused on mechanical design and open-loop control (gait generation). However, in order for such devices to be able to operate in the complex environments for which they are intended, they require exteroceptive sensing to close the ARTICLE IN PRESS www.elsevier.com/locate/neucom 0925-2312/$ - see front matter r 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.neucom.2006.10.139 à Corresponding author. Tel.: +30 2810 391708; fax: +30 2810 391601. E-mail address: tsakiris@ics.forth.gr (D.P. Tsakiris). Please cite this article as: M. Sfakiotakis, D. Tsakiris, Neuromuscular control of reactive behaviors for undulatory robots, Neurocomputing (2007), doi:10.1016/j.neucom.2006.10.139